On 2007-03-23, Ralf Hildebrandt <[email protected]> wrote:
> Grant Edwards wrote:
>> uint32_t r;
>>
>> void foo(uint16_t u1, uint16_t u2)
>> {
>> r = (uint32_t)u1 * (uint32_t)u2;
>> }
>
> This will result in a 32x32 Bit Multiplication with the result
> cropped to 32 Bits.
Which is identical to a 16x16 multiply with a 32 bit result.
> Try:
>
> r = (uint32_t)u1 * u2;
GCC generates the exact same code (as it should).
> Smart compilers will then recognize that you are demanding a
> 16x16=>32Bit multiplication.
I know.
> This is a common problem of C-compilers and it is said that
> most of them recognize it this way.
GCC did recognize it. That's why it called umulhisi3 instead
of umulsi3. umulhisi3 is 16x16->32, umulsi3 is 32x32.
Why does umulhisi3 require 32-bit operands?
--
Grant Edwards grante Yow! ... I want a COLOR
at T.V. and a VIBRATING BED!!!
visi.com
